My organization is considering an RFP for MAID storage and we''re wondering about potential conflicts between MAID and ZFS. We want MAID''s power management benefits but are concerned that what we understand to be ZFS''s use of dynamic striping across devices with filesystem metadata replication and cache syncing will tend to keep disks spinning that the MAID is trying to spin down. Of course, we like ZFS''s large namespace and dynamic memory pool resizing ability. Is it possible to configure ZFS to maximize the benefits of MAID? -John =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-John A. Kunze jak at ucop.edu California Digital Library Work: +1-510-987-9231 415 20th St, #406 http://dot.ucop.edu/home/jak/ Oakland, CA 94612 USA University of California =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- This message posted from opensolaris.org
I think most MAID is sold as a (misguided IMHO) replacement for Tape, not as a Tier 1 kind of storage. YMMV. -- mark John Kunze wrote:> My organization is considering an RFP for MAID storage and we''re > wondering about potential conflicts between MAID and ZFS. > > We want MAID''s power management benefits but are concerned > that what we understand to be ZFS''s use of dynamic striping across > devices with filesystem metadata replication and cache syncing will > tend to keep disks spinning that the MAID is trying to spin down. > Of course, we like ZFS''s large namespace and dynamic memory > pool resizing ability. > > Is it possible to configure ZFS to maximize the benefits of MAID? > > -John > > =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-> John A. Kunze jak at ucop.edu > California Digital Library Work: +1-510-987-9231 > 415 20th St, #406 http://dot.ucop.edu/home/jak/ > Oakland, CA 94612 USA University of California > =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-> > > This message posted from opensolaris.org > _______________________________________________ > zfs-discuss mailing list > zfs-discuss at opensolaris.org > http://mail.opensolaris.org/mailman/listinfo/zfs-discuss >
On Fri, Jun 6, 2008 at 9:29 AM, John Kunze <jak at ucop.edu> wrote:> My organization is considering an RFP for MAID storage and we''re > wondering about potential conflicts between MAID and ZFS.I had to look up MAID, first link Google gave me was http://www.closetmaid.com/ which doesn''t seem right ... MAID seems to be a form of HSM, using powered down disk rather than tape for the offline data. I''ve had poor experience with HSM solutions in the past (only using 1/4 the capacity of tapes, not repacking onto fewer tapes, losing indexes, etc.) but that was several years ago and things may have improved. (I think it was Legato''s product running under Linux, but I''m not certain.) I can''t think of any reason that something like this wouldn''t work with ZFS, though the ACLs may not get saved. -B -- Brandon High bhigh at freaks.com "The good is the enemy of the best." - Nietzsche
Hi ; If you want to use ZFS special ability to pool all the storage together to supply thin provisioning like functionlaty , this will work against MAID. However there is always the option to setup ZFS just like any other FS. (ie. One disk - one fs ) By the way, if I am not mistaken MAID like functionality is built into Solaris. I think solaris gurus should answer this part but I think there is a command to enable MAID like functionality on sata drives. Mertol Mertol Ozyoney Storage Practice - Sales Manager Sun Microsystems, TR Istanbul TR Phone +902123352200 Mobile +905339310752 Fax +902123352222 Email mertol.ozyoney at Sun.COM -----Original Message----- From: zfs-discuss-bounces at opensolaris.org [mailto:zfs-discuss-bounces at opensolaris.org] On Behalf Of John Kunze Sent: Friday, June 06, 2008 7:29 PM To: zfs-discuss at opensolaris.org Subject: [zfs-discuss] ZFS conflict with MAID? My organization is considering an RFP for MAID storage and we''re wondering about potential conflicts between MAID and ZFS. We want MAID''s power management benefits but are concerned that what we understand to be ZFS''s use of dynamic striping across devices with filesystem metadata replication and cache syncing will tend to keep disks spinning that the MAID is trying to spin down. Of course, we like ZFS''s large namespace and dynamic memory pool resizing ability. Is it possible to configure ZFS to maximize the benefits of MAID? -John =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-John A. Kunze jak at ucop.edu California Digital Library Work: +1-510-987-9231 415 20th St, #406 http://dot.ucop.edu/home/jak/ Oakland, CA 94612 USA University of California =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- This message posted from opensolaris.org _______________________________________________ zfs-discuss mailing list zfs-discuss at opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
Hi John, I''ve done some tests with a SUN X4500 with zfs and "MAID" using the powerd of Solaris 10 to power down the disks which weren''t access for a configured time. It''s working fine... The only thing I run into was the problem that it took roundabout a minute to power on 4 disks in a zfs-pool. The problem seems to be that the powerd starts the disks sequentially. I tried to open a RFE... but until now without success. kind regards, Tobias Exner eo ipso Systeme GmbH Mertol Ozyoney schrieb: Hi ; If you want to use ZFS special ability to pool all the storage together to supply thin provisioning like functionlaty , this will work against MAID. However there is always the option to setup ZFS just like any other FS. (ie. One disk - one fs ) By the way, if I am not mistaken MAID like functionality is built into Solaris. I think solaris gurus should answer this part but I think there is a command to enable MAID like functionality on sata drives. Mertol Mertol Ozyoney Storage Practice - Sales Manager Sun Microsystems, TR Istanbul TR Phone +902123352200 Mobile +905339310752 Fax +902123352222 Email mertol.ozyoney@Sun.COM -----Original Message----- From: zfs-discuss-bounces@opensolaris.org [mailto:zfs-discuss-bounces@opensolaris.org] On Behalf Of John Kunze Sent: Friday, June 06, 2008 7:29 PM To: zfs-discuss@opensolaris.org Subject: [zfs-discuss] ZFS conflict with MAID? My organization is considering an RFP for MAID storage and we''re wondering about potential conflicts between MAID and ZFS. We want MAID''s power management benefits but are concerned that what we understand to be ZFS''s use of dynamic striping across devices with filesystem metadata replication and cache syncing will tend to keep disks spinning that the MAID is trying to spin down. Of course, we like ZFS''s large namespace and dynamic memory pool resizing ability. Is it possible to configure ZFS to maximize the benefits of MAID? -John =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-John A. Kunze jak@ucop.edu California Digital Library Work: +1-510-987-9231 415 20th St, #406 http://dot.ucop.edu/home/jak/ Oakland, CA 94612 USA University of California =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- This message posted from opensolaris.org _______________________________________________ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss _______________________________________________ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss _______________________________________________ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
Tobias Exner wrote:> Hi John, > > I''ve done some tests with a SUN X4500 with zfs and "MAID" using the > powerd of Solaris 10 to power down the disks which weren''t access for > a configured time. It''s working fine... > > The only thing I run into was the problem that it took roundabout a > minute to power on 4 disks in a zfs-pool. The problem seems to be that > the powerd starts the disks sequentially.Did you power down disks or spin down disks? It is relatively easy to spin down (or up) disks with luxadm stop (start). If a disk is accessed, then it will spin itself up. By default, the timeout for disk response is 60 seconds, and most disks can spin up in less than 60 seconds.> I tried to open a RFE... but until now without success. >Perhaps because disks will spin up when an access is requested, so to solve your "problem" you''d have to make sure that all of a set of disks are accessed when any in the set are accessed -- butugly. NB. back when I had a largish pile of smallish disks hanging off my workstation for testing, a simple cron job running luxadm stop helped my energy bill :-) -- richard
Richard Elling wrote:> Tobias Exner wrote: > >> Hi John, >> >> I''ve done some tests with a SUN X4500 with zfs and "MAID" using the >> powerd of Solaris 10 to power down the disks which weren''t access for >> a configured time. It''s working fine... >> >> The only thing I run into was the problem that it took roundabout a >> minute to power on 4 disks in a zfs-pool. The problem seems to be that >> the powerd starts the disks sequentially. >> > > Did you power down disks or spin down disks? It is relatively > easy to spin down (or up) disks with luxadm stop (start). If a > disk is accessed, then it will spin itself up. By default, the timeout > for disk response is 60 seconds, and most disks can spin up in > less than 60 seconds.However, some apps will probably be very unhappy if i/o takes 60 seconds to complete.
On Tue, Jun 10, 2008 at 05:32:21PM -0400, Torrey McMahon wrote:> However, some apps will probably be very unhappy if i/o takes 60 seconds > to complete.It''s certainly not uncommon for that to occur in an NFS environment. All of our applications seem to hang on just fine for minor planned and unplanned outages. Would the apps behave differently in this case? (I''m certainly not thinking of a production database for such a configuration). -- Darren
Richard Elling schrieb:> Tobias Exner wrote: >> Hi John, >> >> I''ve done some tests with a SUN X4500 with zfs and "MAID" using the >> powerd of Solaris 10 to power down the disks which weren''t access for >> a configured time. It''s working fine... >> >> The only thing I run into was the problem that it took roundabout a >> minute to power on 4 disks in a zfs-pool. The problem seems to be >> that the powerd starts the disks sequentially. > > Did you power down disks or spin down disks? It is relatively > easy to spin down (or up) disks with luxadm stop (start). If a > disk is accessed, then it will spin itself up. By default, the timeout > for disk response is 60 seconds, and most disks can spin up in > less than 60 seconds.luxadm is not very helpful when I want to have a automatic MAID-solution. The powerd of Solaris just spin down automatically the disks and the powerconsumption falls below 1 watts. ( 3,5") My tests show me that it will take roundabout 20 seconds to power up one single disk and to get access. Actually I don''t know why it takes 55 seconds to spin up 4 disks in a zfs-pool, but that are my results..> >> I tried to open a RFE... but until now without success. >> > > Perhaps because disks will spin up when an access is requested, > so to solve your "problem" you''d have to make sure that all of > a set of disks are accessed when any in the set are accessed -- > butugly.As I know, when I''m using a zfs-pool I have no possibilities to change the behavior which disk will be accessed when I just try to read or write. Do you know more?> > NB. back when I had a largish pile of smallish disks hanging > off my workstation for testing, a simple cron job running > luxadm stop helped my energy bill :-) > -- richard > >regards, Tobias
A Darren Dunham wrote:> On Tue, Jun 10, 2008 at 05:32:21PM -0400, Torrey McMahon wrote: > >> However, some apps will probably be very unhappy if i/o takes 60 seconds >> to complete. >> > > It''s certainly not uncommon for that to occur in an NFS environment. > All of our applications seem to hang on just fine for minor planned and > unplanned outages. > > Would the apps behave differently in this case? (I''m certainly not > thinking of a production database for such a configuration).Some applications have their own internal timers that track i/o time and, if it doesn''t complete in time, will error out. I don''t know which part of the stack the timer was in but I''ve seen an Oracle RAC cluster on QFS timeout much faster then the SCSI retries normally allow for. (I think it was Oracle in that case...)
Torrey McMahon wrote:> A Darren Dunham wrote: > >> On Tue, Jun 10, 2008 at 05:32:21PM -0400, Torrey McMahon wrote: >> >> >>> However, some apps will probably be very unhappy if i/o takes 60 seconds >>> to complete. >>> >>> >> It''s certainly not uncommon for that to occur in an NFS environment. >> All of our applications seem to hang on just fine for minor planned and >> unplanned outages. >> >> Would the apps behave differently in this case? (I''m certainly not >> thinking of a production database for such a configuration). >> > > Some applications have their own internal timers that track i/o time > and, if it doesn''t complete in time, will error out. I don''t know which > part of the stack the timer was in but I''ve seen an Oracle RAC cluster > on QFS timeout much faster then the SCSI retries normally allow for. (I > think it was Oracle in that case...)Oracle bails out after 10 minutes (ORA-27062) ask me how I know... :-P -- richard
Hello Richard,
Thursday, June 12, 2008, 6:54:29 AM, you wrote:
RE> Oracle bails out after 10 minutes (ORA-27062) ask me how I know... :-P
So how do you know?
--
Best regards,
Robert Milkowski mailto:milek at task.gda.pl
http://milek.blogspot.com